Cost feedback for code that agents write
Coding agents now write much of the code that changes your infrastructure, and they open pull requests faster than anyone can check what each one costs. An agent improves with quick feedback, and the bill gives none for a day or more. qloudpulse collects every hour, so a change merged this morning shows in the cost by the afternoon, while the agent, or whoever reviews its work, can still act on it.
Why the bill is too slow
AWS delivers the bill about a day late, and keeps revising it until the month closes. A person who made one change a week could wait for it. An agent that opens ten pull requests a day cannot: by the time the bill shows a step up, several more changes have landed on top of it, and nobody can say which one did it.
What an agent needs is what a test suite gives it: an answer soon after the change, tied to that change. qloudpulse prices what is running every hour, at AWS list price, and sets each change beside the hour it landed.
One change, start to finish
The same loop for a change an agent wrote and one a person wrote. Nothing in it needs the agent to have access to qloudpulse: your pipeline sends the change, and the answers come back where your team and your agents already look.
-
An agent opens a pull request
Infrastructure code, a Helm value, an instance size, a new queue: anything that changes what runs.
-
The pull request says what it will cost
A comment priced at your own rates, from what your resources really use, before anything is applied.
-
It is merged, deployed and marked
The pipeline that deploys sends a change event with the commit, the version and a link to the pull request.
-
The next hourly run prices it
The cost charts mark the change at its hour, and the headline compares the cost with a day and a week before.
-
A jump sends an alert
Within about half an hour of the hour ending, by email, Slack or PagerDuty, naming the resources that rose most.
-
A day, a week and a month later
What the change cost, before against after, for the resources it touched.
Mark each change
A change event is one HTTPS call from the pipeline that deploys: a title, and if you like the service, environment, version and a link. Your company's admins add a source for each sender, such as GitHub Actions, GitLab CI or Argo CD, under Settings, Events, and each source gets its own token. The page gives a ready snippet for each of those, a Check that reads a body as the endpoint would, and a list of what the source last sent.
curl -fsS https://api.qloudpulse.com/v1/events \
-H "Authorization: Bearer $QLOUDPULSE_EVENTS_TOKEN" \
-H "Content-Type: application/json" \
-d '{"type": "deploy", "title": "checkout v2.31.1", "service": "checkout",
"environment": "prod", "version": "v2.31.1"}'
- On the chart. Cost over time draws a marker at each event's hour, with its title and a link to its pull request or release. On a chart cut to one service, only that service's changes are drawn, matched by your own cost tags.
- In the ledger. Under a report's headline, the change ledger lists the changes between the two points it compares, below the costs that moved.
- A retry adds nothing twice. Give each event your own id, such as the pipeline's run id, and a retry replaces it.
Events go to the qloudpulse API, which also takes estimates and is where you will read your data back.
Change events add nothing to your AWS bill. A token is shown once, kept only as a hash, and stops working the moment you rotate or remove its source.
See it within the hour
Every hour the collector lists what is running in your AWS accounts and clusters and prices it. A change that started four more instances, or moved a database to a larger class, shows in the next run.
- The headline. How the cost per hour moved since 1, 8 or 24 hours or 7 days ago, and which resources moved it, such as a fleet going from 4 to 12 instances.
- Why cost changed. A guided walk from 6 hours to 12 months: how much more, where, what kind of change, and how we know.
- Alerts. When a service's cost in one account rises past your thresholds against the same hour a day and a week before, an email, a Slack message or a PagerDuty alert goes out within about half an hour, naming the three resources that rose most. A drop never alerts.
The run rate is at AWS list price, excluding VAT. Data transfer, snapshots and request charges cannot be measured per hour, so they come from the bill if you connect it.
Priced before it is applied
The cheapest moment to catch an expensive change is before it runs. A small command, qloudpulse estimate, reads the plan of a change and prints a comment for the pull request: what the change adds or saves a month, resource by resource, at the rates your company actually pays and from what your existing resources really use, not a guess.
pulumi preview --json -s payments-prod | qloudpulse estimate > comment.md
Your CI posts the comment with the tool it already has, so the command holds no GitHub or GitLab token. The agent that opened the pull request reads it like any other review comment, and can change course before anything is applied. Pulumi previews are priced today; Terraform and OpenTofu plans follow.
Followed afterwards
A change event that lists the resources it touched is followed to the cost it caused. It gets a first reading a day after the change, from the run rate, and then readings at 7 and 30 days from the bill, with discounts and commitments taken off. Each sets the week before the change against the days after it, and says which figures it read.
A new Changes page lists them, linked from the chart's marker and the change ledger, and an alert says once, at the 7-day reading, when a change raised cost. Next to the pull request's estimate, it tells you how close the estimate was.
Questions
Does the agent need access to qloudpulse?
No. Your pipeline sends the change event with its own token, and qloudpulse reads your accounts through the same read-only role as always. What comes back, the chart, the alert and the pull request comment, lands where the agent and your team already work.
Does it matter who wrote the change?
No. A change from an agent and one from a person are marked, priced and followed the same way. The loop is simply most useful where changes come fastest.
What does it cost?
Change events, the run rate and alerts are part of every plan. The pricing page has the plans, and what qloudpulse can add to your own AWS bill if you turn it on.